From ELIZA to ChatGPT: The Desire to Communicate with Machines

1. Origins: Eliza, the First “Therapist” of the 20th Century

In the 1960s, at MIT, Professor Joseph Weizenbaum developed Eliza, considered the first chatbot in history. This program simulated a Rogerian therapist. Its technique was based on pattern matching and phrase substitution. However, it merely returned questions or reworded statements with no real comprehension. Despite its simplicity, Eliza achieved unexpected results: many users believed it truly understood them. This phenomenon was later named the “Eliza Effect” (L.I.A., 2024a).

Eliza proved that a linguistic trick was enough to create the illusion of intelligence. This marked a key moment in AI history. While it didn’t showcase genuine understanding, it sparked ethical and philosophical debates—especially due to the emotional reactions it provoked (Tarnoff, 2023).

2. When Chatbots Began to Remember

2.1 Parry, the Paranoid Schizophrenic Chatbot

After Eliza, the 1970s brought chatbots like Parry. Parry simulated paranoid patterns. These bots had basic memory and context recognition, albeit rudimentary. They also incorporated more complex responses based on previous interactions. Parry, in particular, featured:

• Memory (maintaining an internal state — its paranoid personality).
• Rudimentary context recognition (sensitive topic activation, emotional state adjustment based on user interaction).
• Complex responses based on previous inputs (tailored replies that reflected its paranoid beliefs and emotional shifts) (L.I.A., 2024b).

Interestingly, even psychiatrists couldn’t distinguish Parry from a real human patient. In fact, it could be argued that Parry was the first chatbot to pass a mini version of the Turing Test. Another fun fact: in 1972, Parry participated in a therapy session with Eliza (Khullar, 2023). A true digital celebrity meet-up.

2.2 Jabberwacky, the Fun Chatbot

In the 1990s, programs like Jabberwacky aimed for less rigid interactions with early learning capabilities. Though still rudimentary compared to today’s models, they were a clear step forward from rule-based systems. Jabberwacky’s standout improvement was its learning model—it didn’t need pre-programmed responses; instead, it learned from conversations with humans (Fryer, 2006; L.I.A., 2024a).

Jabberwacky still used patterns like Eliza but no longer depended on fixed responses. It also used working memory, which allowed it to respond immediately—even without full understanding. For example, if a user said, “I feel happy today,” it might reply, “What makes you happy?” This ability to reference previous messages enabled a more natural, human-like flow of conversation. Unlike earlier systems that treated each sentence in isolation, Jabberwacky introduced continuity (Vadrevu, 2018).

A curious fact about Jabberwacky is that some sources cite its creation date as 1988 or 1997. Both are accurate: the project began in 1988 and was publicly released in 1997 (Wikipedia Eng., 2025). Its creator, programmer Rollo Carpenter, designed it with one goal: to be fun (Arya, 2019).

3. A Paradigm Shift: The Rise of GPT Models

The real breakthrough came in 2018 with the release of the first GPT‑1 model. This generative pre-trained model used transformers and vast amounts of data to predict the next word in a sentence (Marr, 2023).

This was followed by GPT‑2 (2019) and GPT‑3 (2020), both capable of generating coherent, varied paragraphs. In November 2022, OpenAI launched ChatGPT, based on GPT‑3.5, as a public chatbot. It could carry conversations, answer questions, and reflect human values (Iffort, 2023; Roumeliotis & Tselikas, 2023).

4. ChatGPT: A Multimodal Conversational Platform

GPT‑4 launched in March 2023. Among its benefits were image understanding and advanced reasoning. Almost a year later, in May 2024, GPT‑4o was released. The “o” stands for omni. This model combines text, voice, image, and audio. It offers a smooth, multimodal experience through ChatGPT (Iffort, 2023).

Recently, OpenAI introduced autonomous agents based on ChatGPT. These agents can connect with visual browsers, terminals, APIs, and other tools. They can also perform complex tasks—all from a conversational interface (Martín Barbero, 2025).

5. Key Differences and Impact

Each historical chatbot has unique features that made them pioneering models. Older chatbots used patterns and scripts, while modern ones rely on neural networks and transformers.

6. What Comes After ChatGPT?

GPT‑5 is expected in 2025, aiming to unify models, tools, and capabilities into one “unified AI” (Disotto, 2025). Meanwhile, competitors like Google’s Gemini or Inflection AI’s Pi are also on the rise. Their releases were delayed for various reasons but are now gaining ground (Okemwa, 2025; Pinzón, 2023).

At the same time, there’s renewed interest in therapy-focused chatbots, like Woebot, which combine AI and psychology (Khullar, 2023).

7. Conclusion

The history of chatbots is the history of humanity’s longing to be understood by machines. This desire has slowly taken shape, as seen in the advanced capabilities of the latest ChatGPT models.

Since their beginnings, chatbots have stirred controversy—especially regarding emotional interactions, a core trait of human nature. This issue was made evident by the Eliza Effect, where people believed they were speaking with a real person, even though Eliza only repeated preset phrases without true understanding.

Despite these early limitations—and perhaps because of them—chatbots have continued to evolve. Today, they support daily tasks, enhance productivity, and even participate in creative processes.

It’s up to us to better understand this technology. Only then can we use it effectively as a tool for support, companionship, or collaboration. And we must always remember that we come from different natures: one biological, the other artificial. These differences shape radically distinct experiences that limit full empathy—though it may appear otherwise. Yet this also happens among humans, causing similar challenges when it comes to collaboration, support, or love. In essence, whenever interaction takes place.

Explore the history of conversational AI and spread the word!